{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"I couldn't easily try Upernet-swin-V2 without using mmsegmentation, so I combined the scripts from the following repository. I hope it would help.\n\n#https://github.com/yassouali/pytorch-segmentation/blob/master/models/upernet.py\n#https://github.com/qubvel/segmentation_models.pytorch/tree/master/segmentation_models_pytorch\n#https://github.com/microsoft/Swin-Transformer/blob/main/models/swin_transformer_v2.py\n\nIf you want to use swin-V1, you can replace all the code part of swin-V2 and it will work.","metadata":{}},{"cell_type":"markdown","source":"I wasn't sure how to accept an image input size other than a multiple of 256. If you want to try it, you can refer to his repository.\n\n#https://github.com/yujiariyasu/siim_covid19_detection/blob/main/ian-siim/classify/skp/models/swin_dlv3p.py","metadata":{}},{"cell_type":"markdown","source":"swinv2-unet's \"decoder_channels=(512,256,128,64)\"  may need to be changed to the appropriate one.","metadata":{}},{"cell_type":"code","source":"model_urls = {\n    \"swinv2_tiny_window16_256\": \"../input/swinv2w/swinv2_tiny_patch4_window8_256.pth\",\n    \"swinv2_small_window8_256\": \"../input/swinv2w/swinv2_small_patch4_window8_256.pth\",\n    \"swinv2_small_window16_256\": \"../input/swinv2w/swinv2_small_patch4_window16_256.pth\",\n    \"swinv2_base_window16_256\": \"../input/swinv2w/swinv2_base_patch4_window16_256.pth\",\n\n}","metadata":{"execution":{"iopub.status.busy":"2022-06-26T06:39:47.988114Z","iopub.execute_input":"2022-06-26T06:39:47.989113Z","iopub.status.idle":"2022-06-26T06:39:48.017981Z","shell.execute_reply.started":"2022-06-26T06:39:47.988518Z","shell.execute_reply":"2022-06-26T06:39:48.016968Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Swin V2 +Unet/Upernet","metadata":{}},{"cell_type":"code","source":"!pip install segmentation_models_pytorch","metadata":{"execution":{"iopub.status.busy":"2022-06-26T06:39:48.038103Z","iopub.execute_input":"2022-06-26T06:39:48.039083Z","iopub.status.idle":"2022-06-26T06:39:58.909315Z","shell.execute_reply.started":"2022-06-26T06:39:48.039037Z","shell.execute_reply":"2022-06-26T06:39:58.907558Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"\n\nfrom re import X\nimport torch,timm\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom torchvision import models\n\n#from utils.helpers import initialize_weights\nfrom itertools import chain\n\n# --------------------------------------------------------\n# Swin Transformer V2\n# Copyright (c) 2022 Microsoft\n# Licensed under The MIT License [see LICENSE for details]\n# Written by Ze Liu\n# --------------------------------------------------------\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torch.utils.checkpoint as checkpoint\nfrom timm.models.layers import DropPath, to_2tuple, trunc_normal_\nimport numpy as np\n\n\nclass Mlp(nn.Module):\n    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):\n        super().__init__()\n        out_features = out_features or in_features\n        hidden_features = hidden_features or in_features\n        self.fc1 = nn.Linear(in_features, hidden_features)\n        self.act = act_layer()\n        self.fc2 = nn.Linear(hidden_features, out_features)\n        self.drop = nn.Dropout(drop)\n\n    def forward(self, x):\n        x = self.fc1(x)\n        x = self.act(x)\n        x = self.drop(x)\n        x = self.fc2(x)\n        x = self.drop(x)\n        return x\n\n\ndef window_partition(x, window_size):\n    \"\"\"\n    Args:\n        x: (B, H, W, C)\n        window_size (int): window size\n    Returns:\n        windows: (num_windows*B, window_size, window_size, C)\n    \"\"\"\n    \n    B, H, W, C = x.shape\n    x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)\n    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)\n    return windows\n\n\ndef window_reverse(windows, window_size, H, W):\n    \"\"\"\n    Args:\n        windows: (num_windows*B, window_size, window_size, C)\n        window_size (int): Window size\n        H (int): Height of image\n        W (int): Width of image\n    Returns:\n        x: (B, H, W, C)\n    \"\"\"\n    B = int(windows.shape[0] / (H * W / window_size / window_size))\n    x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)\n    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)\n    return x\n\n\nclass WindowAttention(nn.Module):\n    r\"\"\" Window based multi-head self attention (W-MSA) module with relative position bias.\n    It supports both of shifted and non-shifted window.\n    Args:\n        dim (int): Number of input channels.\n        window_size (tuple[int]): The height and width of the window.\n        num_heads (int): Number of attention heads.\n        qkv_bias (bool, optional):  If True, add a learnable bias to query, key, value. Default: True\n        attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0\n        proj_drop (float, optional): Dropout ratio of output. Default: 0.0\n        pretrained_window_size (tuple[int]): The height and width of the window in pre-training.\n    \"\"\"\n\n    def __init__(self, dim, window_size, num_heads, qkv_bias=True, attn_drop=0., proj_drop=0.,\n                 pretrained_window_size=[0, 0]):\n\n        super().__init__()\n        self.dim = dim\n        self.window_size = window_size  # Wh, Ww\n        self.pretrained_window_size = pretrained_window_size\n        self.num_heads = num_heads\n\n        self.logit_scale = nn.Parameter(torch.log(10 * torch.ones((num_heads, 1, 1))), requires_grad=True)\n\n        # mlp to generate continuous relative position bias\n        self.cpb_mlp = nn.Sequential(nn.Linear(2, 512, bias=True),\n                                     nn.ReLU(inplace=True),\n                                     nn.Linear(512, num_heads, bias=False))\n\n        # get relative_coords_table\n        relative_coords_h = torch.arange(-(self.window_size[0] - 1), self.window_size[0], dtype=torch.float32)\n        relative_coords_w = torch.arange(-(self.window_size[1] - 1), self.window_size[1], dtype=torch.float32)\n        relative_coords_table = torch.stack(\n            torch.meshgrid([relative_coords_h,\n                            relative_coords_w])).permute(1, 2, 0).contiguous().unsqueeze(0)  # 1, 2*Wh-1, 2*Ww-1, 2\n        if pretrained_window_size[0] > 0:\n            relative_coords_table[:, :, :, 0] /= (pretrained_window_size[0] - 1)\n            relative_coords_table[:, :, :, 1] /= (pretrained_window_size[1] - 1)\n        else:\n            relative_coords_table[:, :, :, 0] /= (self.window_size[0] - 1)\n            relative_coords_table[:, :, :, 1] /= (self.window_size[1] - 1)\n        relative_coords_table *= 8  # normalize to -8, 8\n        relative_coords_table = torch.sign(relative_coords_table) * torch.log2(\n            torch.abs(relative_coords_table) + 1.0) / np.log2(8)\n\n        self.register_buffer(\"relative_coords_table\", relative_coords_table)\n\n        # get pair-wise relative position index for each token inside the window\n        coords_h = torch.arange(self.window_size[0])\n        coords_w = torch.arange(self.window_size[1])\n        coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww\n        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww\n        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww\n        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2\n        relative_coords[:, :, 0] += self.window_size[0] - 1  # shift to start from 0\n        relative_coords[:, :, 1] += self.window_size[1] - 1\n        relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1\n        relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww\n        self.register_buffer(\"relative_position_index\", relative_position_index)\n\n        self.qkv = nn.Linear(dim, dim * 3, bias=False)\n        if qkv_bias:\n            self.q_bias = nn.Parameter(torch.zeros(dim))\n            self.v_bias = nn.Parameter(torch.zeros(dim))\n        else:\n            self.q_bias = None\n            self.v_bias = None\n        self.attn_drop = nn.Dropout(attn_drop)\n        self.proj = nn.Linear(dim, dim)\n        self.proj_drop = nn.Dropout(proj_drop)\n        self.softmax = nn.Softmax(dim=-1)\n\n    def forward(self, x, mask=None):\n        \"\"\"\n        Args:\n            x: input features with shape of (num_windows*B, N, C)\n            mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None\n        \"\"\"\n        B_, N, C = x.shape\n        qkv_bias = None\n        if self.q_bias is not None:\n            qkv_bias = torch.cat((self.q_bias, torch.zeros_like(self.v_bias, requires_grad=False), self.v_bias))\n        qkv = F.linear(input=x, weight=self.qkv.weight, bias=qkv_bias)\n        qkv = qkv.reshape(B_, N, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)\n        q, k, v = qkv[0], qkv[1], qkv[2]  # make torchscript happy (cannot use tensor as tuple)\n\n        # cosine attention\n        attn = (F.normalize(q, dim=-1) @ F.normalize(k, dim=-1).transpose(-2, -1))\n        logit_scale = torch.clamp(self.logit_scale, max=torch.log(torch.tensor(1. / 0.01))).exp()\n        attn = attn * logit_scale\n\n        relative_position_bias_table = self.cpb_mlp(self.relative_coords_table).view(-1, self.num_heads)\n        relative_position_bias = relative_position_bias_table[self.relative_position_index.view(-1)].view(\n            self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1)  # Wh*Ww,Wh*Ww,nH\n        relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww\n        relative_position_bias = 16 * torch.sigmoid(relative_position_bias)\n        attn = attn + relative_position_bias.unsqueeze(0)\n\n        if mask is not None:\n            nW = mask.shape[0]\n            attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)\n            attn = attn.view(-1, self.num_heads, N, N)\n            attn = self.softmax(attn)\n        else:\n            attn = self.softmax(attn)\n\n        attn = self.attn_drop(attn)\n\n        x = (attn @ v).transpose(1, 2).reshape(B_, N, C)\n        x = self.proj(x)\n        x = self.proj_drop(x)\n        return x\n\n    def extra_repr(self) -> str:\n        return f'dim={self.dim}, window_size={self.window_size}, ' \\\n               f'pretrained_window_size={self.pretrained_window_size}, num_heads={self.num_heads}'\n\n    def flops(self, N):\n        # calculate flops for 1 window with token length of N\n        flops = 0\n        # qkv = self.qkv(x)\n        flops += N * self.dim * 3 * self.dim\n        # attn = (q @ k.transpose(-2, -1))\n        flops += self.num_heads * N * (self.dim // self.num_heads) * N\n        #  x = (attn @ v)\n        flops += self.num_heads * N * N * (self.dim // self.num_heads)\n        # x = self.proj(x)\n        flops += N * self.dim * self.dim\n        return flops\n\n\nclass SwinTransformerBlock(nn.Module):\n    r\"\"\" Swin Transformer Block.\n    Args:\n        dim (int): Number of input channels.\n        input_resolution (tuple[int]): Input resulotion.\n        num_heads (int): Number of attention heads.\n        window_size (int): Window size.\n        shift_size (int): Shift size for SW-MSA.\n        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.\n        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True\n        drop (float, optional): Dropout rate. Default: 0.0\n        attn_drop (float, optional): Attention dropout rate. Default: 0.0\n        drop_path (float, optional): Stochastic depth rate. Default: 0.0\n        act_layer (nn.Module, optional): Activation layer. Default: nn.GELU\n        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm\n        pretrained_window_size (int): Window size in pre-training.\n    \"\"\"\n\n    def __init__(self, dim, input_resolution, num_heads, window_size=7, shift_size=0,\n                 mlp_ratio=4., qkv_bias=True, drop=0., attn_drop=0., drop_path=0.,\n                 act_layer=nn.GELU, norm_layer=nn.LayerNorm, pretrained_window_size=0):\n        super().__init__()\n        self.dim = dim\n        self.input_resolution = input_resolution\n        self.num_heads = num_heads\n        self.window_size = window_size\n        self.shift_size = shift_size\n        self.mlp_ratio = mlp_ratio\n        if min(self.input_resolution) <= self.window_size:\n            # if window size is larger than input resolution, we don't partition windows\n            self.shift_size = 0\n            self.window_size = min(self.input_resolution)\n        assert 0 <= self.shift_size < self.window_size, \"shift_size must in 0-window_size\"\n\n        self.norm1 = norm_layer(dim)\n        self.attn = WindowAttention(\n            dim, window_size=to_2tuple(self.window_size), num_heads=num_heads,\n            qkv_bias=qkv_bias, attn_drop=attn_drop, proj_drop=drop,\n            pretrained_window_size=to_2tuple(pretrained_window_size))\n\n        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()\n        self.norm2 = norm_layer(dim)\n        mlp_hidden_dim = int(dim * mlp_ratio)\n        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)\n\n        if self.shift_size > 0:\n            # calculate attention mask for SW-MSA\n            H, W = self.input_resolution\n            img_mask = torch.zeros((1, H, W, 1))  # 1 H W 1\n            h_slices = (slice(0, -self.window_size),\n                        slice(-self.window_size, -self.shift_size),\n                        slice(-self.shift_size, None))\n            w_slices = (slice(0, -self.window_size),\n                        slice(-self.window_size, -self.shift_size),\n                        slice(-self.shift_size, None))\n            cnt = 0\n            for h in h_slices:\n                for w in w_slices:\n                    img_mask[:, h, w, :] = cnt\n                    cnt += 1\n\n            mask_windows = window_partition(img_mask, self.window_size)  # nW, window_size, window_size, 1\n            mask_windows = mask_windows.view(-1, self.window_size * self.window_size)\n            attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)\n            attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))\n        else:\n            attn_mask = None\n\n        self.register_buffer(\"attn_mask\", attn_mask)\n\n    def forward(self, x):\n        H, W = self.input_resolution\n        B, L, C = x.shape\n        assert L == H * W, \"input feature has wrong size\"\n\n        shortcut = x\n        x = x.view(B, H, W, C)\n\n        # cyclic shift\n        if self.shift_size > 0:\n            shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))\n        else:\n            shifted_x = x\n\n        # partition windows\n        x_windows = window_partition(shifted_x, self.window_size)  # nW*B, window_size, window_size, C\n        x_windows = x_windows.view(-1, self.window_size * self.window_size, C)  # nW*B, window_size*window_size, C\n\n        # W-MSA/SW-MSA\n        attn_windows = self.attn(x_windows, mask=self.attn_mask)  # nW*B, window_size*window_size, C\n\n        # merge windows\n        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)\n        shifted_x = window_reverse(attn_windows, self.window_size, H, W)  # B H' W' C\n\n        # reverse cyclic shift\n        if self.shift_size > 0:\n            x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))\n        else:\n            x = shifted_x\n        x = x.view(B, H * W, C)\n        x = shortcut + self.drop_path(self.norm1(x))\n\n        # FFN\n        x = x + self.drop_path(self.norm2(self.mlp(x)))\n\n        return x\n\n    def extra_repr(self) -> str:\n        return f\"dim={self.dim}, input_resolution={self.input_resolution}, num_heads={self.num_heads}, \" \\\n               f\"window_size={self.window_size}, shift_size={self.shift_size}, mlp_ratio={self.mlp_ratio}\"\n\n    def flops(self):\n        flops = 0\n        H, W = self.input_resolution\n        # norm1\n        flops += self.dim * H * W\n        # W-MSA/SW-MSA\n        nW = H * W / self.window_size / self.window_size\n        flops += nW * self.attn.flops(self.window_size * self.window_size)\n        # mlp\n        flops += 2 * H * W * self.dim * self.dim * self.mlp_ratio\n        # norm2\n        flops += self.dim * H * W\n        return flops\n\n\nclass PatchMerging(nn.Module):\n    r\"\"\" Patch Merging Layer.\n    Args:\n        input_resolution (tuple[int]): Resolution of input feature.\n        dim (int): Number of input channels.\n        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm\n    \"\"\"\n\n    def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):\n        super().__init__()\n        self.input_resolution = input_resolution\n        self.dim = dim\n        self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)\n        self.norm = norm_layer(2 * dim)\n\n    def forward(self, x):\n        \"\"\"\n        x: B, H*W, C\n        \"\"\"\n        H, W = self.input_resolution\n        B, L, C = x.shape\n        assert L == H * W, \"input feature has wrong size\"\n        assert H % 2 == 0 and W % 2 == 0, f\"x size ({H}*{W}) are not even.\"\n\n        x = x.view(B, H, W, C)\n\n        x0 = x[:, 0::2, 0::2, :]  # B H/2 W/2 C\n        x1 = x[:, 1::2, 0::2, :]  # B H/2 W/2 C\n        x2 = x[:, 0::2, 1::2, :]  # B H/2 W/2 C\n        x3 = x[:, 1::2, 1::2, :]  # B H/2 W/2 C\n        x = torch.cat([x0, x1, x2, x3], -1)  # B H/2 W/2 4*C\n        x = x.view(B, -1, 4 * C)  # B H/2*W/2 4*C\n\n        x = self.reduction(x)\n        x = self.norm(x)\n\n        return x\n\n    def extra_repr(self) -> str:\n        return f\"input_resolution={self.input_resolution}, dim={self.dim}\"\n\n    def flops(self):\n        H, W = self.input_resolution\n        flops = (H // 2) * (W // 2) * 4 * self.dim * 2 * self.dim\n        flops += H * W * self.dim // 2\n        return flops\n\n\nclass BasicLayer(nn.Module):\n    \"\"\" A basic Swin Transformer layer for one stage.\n    Args:\n        dim (int): Number of input channels.\n        input_resolution (tuple[int]): Input resolution.\n        depth (int): Number of blocks.\n        num_heads (int): Number of attention heads.\n        window_size (int): Local window size.\n        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.\n        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True\n        drop (float, optional): Dropout rate. Default: 0.0\n        attn_drop (float, optional): Attention dropout rate. Default: 0.0\n        drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0\n        norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm\n        downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None\n        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.\n        pretrained_window_size (int): Local window size in pre-training.\n    \"\"\"\n\n    def __init__(self, dim, input_resolution, depth, num_heads, window_size,\n                 mlp_ratio=4., qkv_bias=True, drop=0., attn_drop=0.,\n                 drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False,\n                 pretrained_window_size=0):\n\n        super().__init__()\n        self.dim = dim\n        self.input_resolution = input_resolution\n        self.depth = depth\n        self.use_checkpoint = use_checkpoint\n\n        # build blocks\n        self.blocks = nn.ModuleList([\n            SwinTransformerBlock(dim=dim, input_resolution=input_resolution,\n                                 num_heads=num_heads, window_size=window_size,\n                                 shift_size=0 if (i % 2 == 0) else window_size // 2,\n                                 mlp_ratio=mlp_ratio,\n                                 qkv_bias=qkv_bias,\n                                 drop=drop, attn_drop=attn_drop,\n                                 drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,\n                                 norm_layer=norm_layer,\n                                 pretrained_window_size=pretrained_window_size)\n            for i in range(depth)])\n\n        # patch merging layer\n        if downsample is not None:\n            self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer)\n        else:\n            self.downsample = None\n\n    def forward(self, x):\n        for blk in self.blocks:\n            if self.use_checkpoint:\n                x = checkpoint.checkpoint(blk, x)\n            else:\n                x = blk(x)\n        if self.downsample is not None:\n            x = self.downsample(x)\n        return x\n\n    def extra_repr(self) -> str:\n        return f\"dim={self.dim}, input_resolution={self.input_resolution}, depth={self.depth}\"\n\n    def flops(self):\n        flops = 0\n        for blk in self.blocks:\n            flops += blk.flops()\n        if self.downsample is not None:\n            flops += self.downsample.flops()\n        return flops\n\n    def _init_respostnorm(self):\n        for blk in self.blocks:\n            nn.init.constant_(blk.norm1.bias, 0)\n            nn.init.constant_(blk.norm1.weight, 0)\n            nn.init.constant_(blk.norm2.bias, 0)\n            nn.init.constant_(blk.norm2.weight, 0)\n\n\nclass PatchEmbed(nn.Module):\n    r\"\"\" Image to Patch Embedding\n    Args:\n        img_size (int): Image size.  Default: 224.\n        patch_size (int): Patch token size. Default: 4.\n        in_chans (int): Number of input image channels. Default: 3.\n        embed_dim (int): Number of linear projection output channels. Default: 96.\n        norm_layer (nn.Module, optional): Normalization layer. Default: None\n    \"\"\"\n\n    def __init__(self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):\n        super().__init__()\n        img_size = to_2tuple(img_size)\n        patch_size = to_2tuple(patch_size)\n        patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]]\n        self.img_size = img_size\n        self.patch_size = patch_size\n        self.patches_resolution = patches_resolution\n        self.num_patches = patches_resolution[0] * patches_resolution[1]\n\n        self.in_chans = in_chans\n        self.embed_dim = embed_dim\n\n        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)\n        if norm_layer is not None:\n            self.norm = norm_layer(embed_dim)\n        else:\n            self.norm = None\n\n    def forward(self, x):\n        B, C, H, W = x.shape\n        # FIXME look at relaxing size constraints\n        assert H == self.img_size[0] and W == self.img_size[1], \\\n            f\"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]}).\"\n        x = self.proj(x).flatten(2).transpose(1, 2)  # B Ph*Pw C\n        if self.norm is not None:\n            x = self.norm(x)\n        return x\n\n    def flops(self):\n        Ho, Wo = self.patches_resolution\n        flops = Ho * Wo * self.embed_dim * self.in_chans * (self.patch_size[0] * self.patch_size[1])\n        if self.norm is not None:\n            flops += Ho * Wo * self.embed_dim\n        return flops\n\n\nclass SwinTransformerV2(nn.Module):\n    r\"\"\" Swin Transformer\n        A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows`  -\n          https://arxiv.org/pdf/2103.14030\n    Args:\n        img_size (int | tuple(int)): Input image size. Default 224\n        patch_size (int | tuple(int)): Patch size. Default: 4\n        in_chans (int): Number of input image channels. Default: 3\n        num_classes (int): Number of classes for classification head. Default: 1000\n        embed_dim (int): Patch embedding dimension. Default: 96\n        depths (tuple(int)): Depth of each Swin Transformer layer.\n        num_heads (tuple(int)): Number of attention heads in different layers.\n        window_size (int): Window size. Default: 7\n        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4\n        qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True\n        drop_rate (float): Dropout rate. Default: 0\n        attn_drop_rate (float): Attention dropout rate. Default: 0\n        drop_path_rate (float): Stochastic depth rate. Default: 0.1\n        norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm.\n        ape (bool): If True, add absolute position embedding to the patch embedding. Default: False\n        patch_norm (bool): If True, add normalization after patch embedding. Default: True\n        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False\n        pretrained_window_sizes (tuple(int)): Pretrained window sizes of each layer.\n    \"\"\"\n\n    def __init__(self, img_size=224, patch_size=4, in_chans=3, num_classes=1000,\n                 embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24],\n                 window_size=7, mlp_ratio=4., qkv_bias=True,\n                 drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1,\n                 norm_layer=nn.LayerNorm, ape=False, patch_norm=True,\n                 use_checkpoint=False, pretrained_window_sizes=[0, 0, 0, 0], **kwargs):\n        super().__init__()\n\n        self.num_classes = num_classes\n        self.num_layers = len(depths)\n        self.embed_dim = embed_dim\n        self.ape = ape\n        self.patch_norm = patch_norm\n        self.num_features = int(embed_dim * 2 ** (self.num_layers - 1))\n        self.mlp_ratio = mlp_ratio\n\n        # split image into non-overlapping patches\n        self.patch_embed = PatchEmbed(\n            img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim,\n            norm_layer=norm_layer if self.patch_norm else None)\n        num_patches = self.patch_embed.num_patches\n        patches_resolution = self.patch_embed.patches_resolution\n        self.patches_resolution = patches_resolution\n\n        # absolute position embedding\n        if self.ape:\n            self.absolute_pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim))\n            trunc_normal_(self.absolute_pos_embed, std=.02)\n\n        self.pos_drop = nn.Dropout(p=drop_rate)\n\n        # stochastic depth\n        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]  # stochastic depth decay rule\n\n        # build layers\n        self.layers = nn.ModuleList()\n        for i_layer in range(self.num_layers):\n            layer = BasicLayer(dim=int(embed_dim * 2 ** i_layer),\n                               input_resolution=(patches_resolution[0] // (2 ** i_layer),\n                                                 patches_resolution[1] // (2 ** i_layer)),\n                               depth=depths[i_layer],\n                               num_heads=num_heads[i_layer],\n                               window_size=window_size,\n                               mlp_ratio=self.mlp_ratio,\n                               qkv_bias=qkv_bias,\n                               drop=drop_rate, attn_drop=attn_drop_rate,\n                               drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])],\n                               norm_layer=norm_layer,\n                               downsample=PatchMerging if (i_layer < self.num_layers - 1) else None,\n                               use_checkpoint=use_checkpoint,\n                               pretrained_window_size=pretrained_window_sizes[i_layer])\n            self.layers.append(layer)\n\n        self.norm = norm_layer(self.num_features)\n        self.avgpool = nn.AdaptiveAvgPool1d(1)\n        self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity()\n\n        self.apply(self._init_weights)\n        for bly in self.layers:\n            bly._init_respostnorm()\n\n    def _init_weights(self, m):\n        if isinstance(m, nn.Linear):\n            trunc_normal_(m.weight, std=.02)\n            if isinstance(m, nn.Linear) and m.bias is not None:\n                nn.init.constant_(m.bias, 0)\n        elif isinstance(m, nn.LayerNorm):\n            nn.init.constant_(m.bias, 0)\n            nn.init.constant_(m.weight, 1.0)\n\n    @torch.jit.ignore\n    def no_weight_decay(self):\n        return {'absolute_pos_embed'}\n\n    @torch.jit.ignore\n    def no_weight_decay_keywords(self):\n        return {\"cpb_mlp\", \"logit_scale\", 'relative_position_bias_table'}\n\n    def forward_features(self, x):\n        x = self.patch_embed(x)\n        if self.ape:\n            x = x + self.absolute_pos_embed\n        x = self.pos_drop(x)\n\n        for layer in self.layers:\n            x = layer(x)\n\n        x = self.norm(x)  # B L C\n        x = self.avgpool(x.transpose(1, 2))  # B C 1\n        x = torch.flatten(x, 1)\n        return x\n    def extra_features(self,x):\n        x = self.patch_embed(x)\n        if self.ape:\n            x = x + self.absolute_pos_embed\n        x = self.pos_drop(x)\n        feature = []\n        \n        for layer in self.layers:\n            x = layer(x)\n            bs,n,f=x.shape\n            h = int(n**0.5)\n\n            feature.append(x.view(-1,h,h,f).permute(0, 3, 1, 2).contiguous())\n        return feature\n\n    \n    def get_unet_feature(self,x):\n        x = self.patch_embed(x)\n        if self.ape:\n            x = x + self.absolute_pos_embed\n        x = self.pos_drop(x)\n        bs,n,f=x.shape\n        h = int(n**0.5)\n        feature = [x.view(-1,h,h,f).permute(0, 3, 1, 2).contiguous()]\n        \n        for layer in self.layers:\n            x = layer(x)\n            bs,n,f=x.shape\n            h = int(n**0.5)\n\n            feature.append(x.view(-1,h,h,f).permute(0, 3, 1, 2).contiguous())\n        return feature\n\n    def forward(self, x):\n        x = self.forward_features(x)\n        x = self.head(x)\n        return x\n\n\n    def flops(self):\n        flops = 0\n        flops += self.patch_embed.flops()\n        for i, layer in enumerate(self.layers):\n            flops += layer.flops()\n        flops += self.num_features * self.patches_resolution[0] * self.patches_resolution[1] // (2 ** self.num_layers)\n        flops += self.num_features * self.num_classes\n        return flops\n\n\n\ndef swin_v2(size,img_size=256,in_22k=False, **kwargs):\n    if size==\"swinv2_tiny_window16_256\":\n        model = SwinTransformerV2(img_size=img_size,window_size=16,embed_dim=96,depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24], **kwargs)\n        checkpoint=torch.load(model_urls[size])[\"model\"]\n        if img_size!=256:\n            del checkpoint[\"layers.0.blocks.0.attn.relative_coords_table\"]\n            del checkpoint[\"layers.0.blocks.0.attn.relative_position_index\"]\n            del checkpoint[\"layers.0.blocks.1.attn_mask\"]\n            del checkpoint[\"layers.0.blocks.1.attn.relative_coords_table\"]\n            del checkpoint[\"layers.0.blocks.1.attn.relative_position_index\"]\n            del checkpoint[\"layers.1.blocks.0.attn.relative_coords_table\"]\n            del checkpoint[\"layers.1.blocks.0.attn.relative_position_index\"]\n            del checkpoint[\"layers.1.blocks.1.attn_mask\"]\n            del checkpoint[\"layers.1.blocks.1.attn.relative_coords_table\"]\n            del checkpoint[\"layers.1.blocks.1.attn.relative_position_index\"]\n            del checkpoint[\"layers.2.blocks.0.attn.relative_coords_table\"]\n            del checkpoint[\"layers.2.blocks.0.attn.relative_position_index\"]\n            del checkpoint[\"layers.2.blocks.1.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.1.attn.relative_coords_table\"]\n            del checkpoint[\"layers.2.blocks.1.attn.relative_position_index\"]\n            del checkpoint[\"layers.2.blocks.2.attn.relative_coords_table\"]\n            del checkpoint[\"layers.2.blocks.2.attn.relative_position_index\"]\n            del checkpoint[\"layers.2.blocks.3.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.3.attn.relative_coords_table\"]\n            del checkpoint[\"layers.2.blocks.3.attn.relative_position_index\"]\n            del checkpoint[\"layers.2.blocks.4.attn.relative_coords_table\"]\n            del checkpoint[\"layers.2.blocks.4.attn.relative_position_index\"]\n            del checkpoint[\"layers.2.blocks.5.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.5.attn.relative_coords_table\"]\n            del checkpoint[\"layers.2.blocks.5.attn.relative_position_index\"]\n            del checkpoint[\"layers.3.blocks.0.attn.relative_coords_table\"]\n            del checkpoint[\"layers.3.blocks.0.attn.relative_position_index\"]\n            del checkpoint[\"layers.3.blocks.1.attn.relative_coords_table\"]\n            del checkpoint[\"layers.3.blocks.1.attn.relative_position_index\"]\n        model.load_state_dict(checkpoint, strict=False)\n    elif size==\"swinv2_small_window8_256\":\n        model = SwinTransformerV2(img_size=img_size,window_size=8,embed_dim=96,depths=[2, 2, 18, 2], num_heads=[3, 6, 12, 24], **kwargs)\n        checkpoint=torch.load(model_urls[size])[\"model\"]\n        if img_size!=256:\n            del checkpoint[\"layers.0.blocks.1.attn_mask\"]\n            del checkpoint[\"layers.1.blocks.1.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.1.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.3.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.5.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.7.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.9.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.11.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.13.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.15.attn_mask\"]\n            del checkpoint[\"layers.2.blocks.17.attn_mask\"]\n\n        model.load_state_dict(checkpoint, strict=False)\n    elif size==\"swinv2_small_window16_256\":\n        model = SwinTransformerV2(img_size=img_size,window_size=16,embed_dim=96,depths=[2, 2, 18, 2], num_heads=[3, 6, 12, 24], **kwargs)\n        checkpoint=torch.load(model_urls[size])[\"model\"]\n        if img_size!=256:\n            del checkpoint[\"layers.0.blocks.1.attn_mask\"]\n            del checkpoint[\"layers.1.blocks.1.attn_mask\"]\n            del checkpoint[\"layers.3.blocks.0.attn.relative_coords_table\"]\n            del checkpoint[\"layers.3.blocks.0.attn.relative_position_index\"]\n            del checkpoint[\"layers.3.blocks.1.attn.relative_coords_table\"]\n            del checkpoint[\"layers.3.blocks.1.attn.relative_position_index\"]\n        model.load_state_dict(checkpoint, strict=False)\n    elif size==\"swinv2_base_window16_256\":\n        model = SwinTransformerV2(img_size=img_size,window_size=16,embed_dim=128,depths=[2, 2, 18, 2], num_heads=[4, 8, 16, 32], **kwargs)\n        checkpoint=torch.load(model_urls[size])[\"model\"]\n        if img_size!=256:\n            del checkpoint[\"layers.0.blocks.1.attn_mask\"]\n            del checkpoint[\"layers.1.blocks.1.attn_mask\"]\n            del checkpoint[\"layers.3.blocks.0.attn.relative_coords_table\"]\n            del checkpoint[\"layers.3.blocks.0.attn.relative_position_index\"]\n            del checkpoint[\"layers.3.blocks.1.attn.relative_coords_table\"]\n            del checkpoint[\"layers.3.blocks.1.attn.relative_position_index\"]\n        model.load_state_dict(checkpoint, strict=False)\n\n    return model\n\nclass PSPModule(nn.Module):\n    # In the original inmplementation they use precise RoI pooling \n    # Instead of using adaptative average pooling\n    def __init__(self, in_channels, bin_sizes=[1, 2, 4, 6]):\n        super(PSPModule, self).__init__()\n        out_channels = in_channels // len(bin_sizes)\n        self.stages = nn.ModuleList([self._make_stages(in_channels, out_channels, b_s) \n                                                        for b_s in bin_sizes])\n        self.bottleneck = nn.Sequential(\n            nn.Conv2d(in_channels+(out_channels * len(bin_sizes)), in_channels, \n                                    kernel_size=3, padding=1, bias=False),\n            nn.BatchNorm2d(in_channels),\n            nn.ReLU(inplace=True),\n            nn.Dropout2d(0.1)\n        )\n\n    def _make_stages(self, in_channels, out_channels, bin_sz):\n        prior = nn.AdaptiveAvgPool2d(output_size=bin_sz)\n        conv = nn.Conv2d(in_channels, out_channels, kernel_size=1, bias=False)\n        bn = nn.BatchNorm2d(out_channels)\n        relu = nn.ReLU(inplace=True)\n        return nn.Sequential(prior, conv, bn, relu)\n    \n    def forward(self, features):\n        h, w = features.size()[2], features.size()[3]\n        pyramids = [features]\n        pyramids.extend([F.interpolate(stage(features), size=(h, w), mode='bilinear', \n                                        align_corners=True) for stage in self.stages])\n        output = self.bottleneck(torch.cat(pyramids, dim=1))\n        return output\n\nclass ResNet(nn.Module):\n    def __init__(self, in_channels=3, output_stride=16, backbone='resnet101', pretrained=True):\n        super(ResNet, self).__init__()\n        model = getattr(models, backbone)(pretrained)\n        if not pretrained or in_channels != 3:\n            self.initial = nn.Sequential(\n                nn.Conv2d(in_channels, 64, 7, stride=2, padding=3, bias=False),\n                nn.BatchNorm2d(64),\n                nn.ReLU(inplace=True),\n                nn.MaxPool2d(kernel_size=3, stride=2, padding=1)\n            )\n            #initialize_weights(self.initial)\n        else:\n            self.initial = nn.Sequential(*list(model.children())[:4])\n        \n        self.layer1 = model.layer1\n        self.layer2 = model.layer2\n        self.layer3 = model.layer3\n        self.layer4 = model.layer4\n\n        if output_stride == 16: s3, s4, d3, d4 = (2, 1, 1, 2)\n        elif output_stride == 8: s3, s4, d3, d4 = (1, 1, 2, 4)\n\n        if output_stride == 8: \n            for n, m in self.layer3.named_modules():\n                if 'conv1' in n and (backbone == 'resnet34' or backbone == 'resnet18'):\n                    m.dilation, m.padding, m.stride = (d3,d3), (d3,d3), (s3,s3)\n                elif 'conv2' in n:\n                    m.dilation, m.padding, m.stride = (d3,d3), (d3,d3), (s3,s3)\n                elif 'downsample.0' in n:\n                    m.stride = (s3, s3)\n\n        for n, m in self.layer4.named_modules():\n            if 'conv1' in n and (backbone == 'resnet34' or backbone == 'resnet18'):\n                m.dilation, m.padding, m.stride = (d4,d4), (d4,d4), (s4,s4)\n            elif 'conv2' in n:\n                m.dilation, m.padding, m.stride = (d4,d4), (d4,d4), (s4,s4)\n            elif 'downsample.0' in n:\n                m.stride = (s4, s4)\n\n    def forward(self, x):\n        x = self.initial(x)\n        x1 = self.layer1(x)\n        print(\"x1\",x1.shape)\n        x2 = self.layer2(x1)\n        print(\"x2\",x2.shape)\n        x3 = self.layer3(x2)\n        print(\"x3\",x3.shape)\n        x4 = self.layer4(x3)\n        print(\"x4\",x4.shape)\n\n        return [x1, x2, x3, x4]\n\ndef up_and_add(x, y):\n    return F.interpolate(x, size=(y.size(2), y.size(3)), mode='bilinear', align_corners=True) + y\n\nclass FPN_fuse(nn.Module):\n    def __init__(self, feature_channels=[256, 512, 1024, 2048], fpn_out=256):\n        super(FPN_fuse, self).__init__()\n        assert feature_channels[0] == fpn_out\n        self.conv1x1 = nn.ModuleList([nn.Conv2d(ft_size, fpn_out, kernel_size=1)\n                                    for ft_size in feature_channels[1:]])\n        self.smooth_conv =  nn.ModuleList([nn.Conv2d(fpn_out, fpn_out, kernel_size=3, padding=1)] \n                                    * (len(feature_channels)-1))\n        self.conv_fusion = nn.Sequential(\n            nn.Conv2d(len(feature_channels)*fpn_out, fpn_out, kernel_size=3, padding=1, bias=False),\n            nn.BatchNorm2d(fpn_out),\n            nn.ReLU(inplace=True)\n        )\n\n    def forward(self, features):\n        \n        features[1:] = [conv1x1(feature) for feature, conv1x1 in zip(features[1:], self.conv1x1)]##\n        P = [up_and_add(features[i], features[i-1]) for i in reversed(range(1, len(features)))]\n        P = [smooth_conv(x) for smooth_conv, x in zip(self.smooth_conv, P)]\n        P = list(reversed(P))\n        P.append(features[-1]) #P = [P1, P2, P3, P4]\n        H, W = P[0].size(2), P[0].size(3)\n        P[1:] = [F.interpolate(feature, size=(H, W), mode='bilinear', align_corners=True) for feature in P[1:]]\n\n        x = self.conv_fusion(torch.cat((P), dim=1))\n        return x\n\nclass UperNet_swin(nn.Module):\n    # Implementing only the object path\n    def __init__(self,size=\"swinv2_small_window16_256\",img_size=256,num_classes=1, in_channels=3, pretrained=True):\n        super(UperNet_swin, self).__init__()\n\n\n        self.backbone = swin_v2(size=size,img_size=img_size)\n        if size.split(\"_\")[1] in [\"small\",\"tiny\"]:\n            feature_channels = [192,384,768,768]\n        elif size.split(\"_\")[1] in [\"base\"]:\n            feature_channels = [256,512,1024,1024]\n        self.PPN = PSPModule(feature_channels[-1])\n        self.FPN = FPN_fuse(feature_channels, fpn_out=feature_channels[0])\n        self.head = nn.Conv2d(feature_channels[0], num_classes, kernel_size=3, padding=1)\n\n\n\n    def forward(self, x):\n        input_size = (x.size()[2], x.size()[3])\n\n        features = self.backbone.extra_features(x)\n        features[-1] = self.PPN(features[-1])\n        x = self.head(self.FPN(features))\n\n        x = F.interpolate(x, size=input_size, mode='bilinear')\n        return x\n\n    def get_backbone_params(self):\n        return self.backbone.parameters()\n\n    def get_decoder_params(self):\n        return chain(self.PPN.parameters(), self.FPN.parameters(), self.head.parameters())\n\n    def freeze_bn(self):\n        for module in self.modules():\n            if isinstance(module, nn.BatchNorm2d): module.eval()\n\nfrom segmentation_models_pytorch.base import modules as md\n\nfrom  timm.models.layers.cbam import *\n\nclass DecoderBlock(nn.Module):\n    def __init__(\n        self,\n        in_channels,\n        skip_channels,\n        out_channels,\n        use_batchnorm=True,\n        attention_type=None,\n    ):\n        super().__init__()\n        self.conv1 = md.Conv2dReLU(\n            in_channels + skip_channels,\n            out_channels,\n            kernel_size=3,\n            padding=1,\n            use_batchnorm=use_batchnorm,\n        )\n        if attention_type==\"cbam\":\n            self.attention1 = CbamModule(channels=in_channels + skip_channels)\n        else:\n            self.attention1 = md.Attention(attention_type, in_channels=in_channels + skip_channels)\n        self.conv2 = md.Conv2dReLU(\n            out_channels,\n            out_channels,\n            kernel_size=3,\n            padding=1,\n            use_batchnorm=use_batchnorm,\n        )\n        if attention_type==\"cbam\":\n            self.attention2 = CbamModule(channels=out_channels)\n        else:\n            self.attention2 = md.Attention(attention_type, in_channels=out_channels)\n        self.in_channels=in_channels\n        self.out_channels = out_channels\n        self.skip_channels = skip_channels\n    def forward(self, x, skip=None):\n        if skip is None:\n            x = F.interpolate(x, scale_factor=2, mode=\"nearest\")\n        else:\n            if x.shape[-1]!=skip.shape[-1]:\n                x = F.interpolate(x, scale_factor=2, mode=\"nearest\")\n        if skip is not None:\n            #print(x.shape,skip.shape)\n            x = torch.cat([x, skip], dim=1)\n            x = self.attention1(x)\n        x = self.conv1(x)\n        x = self.conv2(x)\n        x = self.attention2(x)\n        return x\n\n\nclass CenterBlock(nn.Sequential):\n    def __init__(self, in_channels, out_channels, use_batchnorm=True):\n        conv1 = md.Conv2dReLU(\n            in_channels,\n            out_channels,\n            kernel_size=3,\n            padding=1,\n            use_batchnorm=use_batchnorm,\n        )\n        conv2 = md.Conv2dReLU(\n            out_channels,\n            out_channels,\n            kernel_size=3,\n            padding=1,\n            use_batchnorm=use_batchnorm,\n        )\n        super().__init__(conv1, conv2)\n\n\nclass UnetDecoder(nn.Module):\n    def __init__(\n        self,\n        encoder_channels,\n        decoder_channels,\n        n_blocks=5,\n        use_batchnorm=True,\n        attention_type=None,\n        center=False,\n    ):\n        super().__init__()\n\n        if n_blocks != len(decoder_channels):\n            raise ValueError(\n                \"Model depth is {}, but you provide `decoder_channels` for {} blocks.\".format(\n                    n_blocks, len(decoder_channels)\n                )\n            )\n\n        # remove first skip with same spatial resolution\n        encoder_channels = encoder_channels[1:]\n        # reverse channels to start from head of encoder\n        encoder_channels = encoder_channels[::-1]\n\n        # computing blocks input and output channels\n        head_channels = encoder_channels[0]\n        in_channels = [head_channels] + list(decoder_channels[:-1])\n        skip_channels = list(encoder_channels[1:]) + [0]\n\n        out_channels = decoder_channels\n\n        if center:\n            self.center = CenterBlock(head_channels, head_channels, use_batchnorm=use_batchnorm)\n        else:\n            self.center = nn.Identity()\n\n        # combine decoder keyword arguments\n        kwargs = dict(use_batchnorm=use_batchnorm, attention_type=attention_type)\n        blocks = [\n            DecoderBlock(in_ch, skip_ch, out_ch, **kwargs)\n            for in_ch, skip_ch, out_ch in zip(in_channels, skip_channels, out_channels)\n        ]\n        self.blocks = nn.ModuleList(blocks)\n\n    def forward(self, *features):\n\n        features = features[1:]  # remove first skip with same spatial resolution\n        features = features[::-1]  # reverse channels to start from head of encoder\n\n        head = features[0]\n        skips = features[1:]\n\n        x = self.center(head)\n        for i, decoder_block in enumerate(self.blocks):\n            skip = skips[i] if i < len(skips) else None\n            x = decoder_block(x, skip)\n            #ここでhypercolumns\n            #y_i = self.upsample1(y_i)\n        #hypercol = torch.cat([y0,y1,y2,y3,y4], dim=1)\n\n        return x\n\nclass SegmentationHead(nn.Sequential):\n    def __init__(self, in_channels, out_channels, kernel_size=3, upsampling=1):\n        conv2d = nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, padding=kernel_size // 2)\n        upsampling = nn.UpsamplingBilinear2d(scale_factor=upsampling) if upsampling > 1 else nn.Identity()\n        super().__init__(conv2d, upsampling)\n\nclass unet_swin(nn.Module):\n\n    def __init__(\n        self,size=\"small\",img_size=256 #\"base\" \"large\"\n    ):\n        super().__init__()\n\n        self.encoder = swin_v2(size=size,img_size=img_size)\n\n        if size.split(\"_\")[1] in [\"small\",\"tiny\"]:\n            feature_channels = (3,192,384,768,768)\n        elif size.split(\"_\")[1] in [\"base\"]:\n            feature_channels = (3,256,512,1024,1024)\n        self.decoder = UnetDecoder(encoder_channels=feature_channels,n_blocks=4,decoder_channels=(512,256,128,64),attention_type=None)\n\n        self.segmentation_head = SegmentationHead(in_channels=64,out_channels=1,kernel_size=3,upsampling=4\n        )\n\n    def forward(self, input):\n        encoder_featrue = self.encoder.get_unet_feature(input)\n        decoder_output = self.decoder(*encoder_featrue)\n        masks = self.segmentation_head(decoder_output)\n\n        return masks\n","metadata":{"execution":{"iopub.status.busy":"2022-06-26T06:39:58.913466Z","iopub.execute_input":"2022-06-26T06:39:58.914078Z","iopub.status.idle":"2022-06-26T06:40:08.271151Z","shell.execute_reply.started":"2022-06-26T06:39:58.914024Z","shell.execute_reply":"2022-06-26T06:40:08.269607Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import time\ns = time.time()\nS = 512\ninp = torch.randn((4,3,S,S))\n\nmodel = unet_swin(img_size=S,size=\"swinv2_base_window16_256\")\nout = model(inp)\n\nprint(out.shape,time.time()-s)\ndel model,out","metadata":{"execution":{"iopub.status.busy":"2022-06-26T06:40:08.273634Z","iopub.execute_input":"2022-06-26T06:40:08.274964Z","iopub.status.idle":"2022-06-26T06:40:29.636560Z","shell.execute_reply.started":"2022-06-26T06:40:08.274909Z","shell.execute_reply":"2022-06-26T06:40:29.634946Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"s = time.time()\nS = 256\ninp = torch.randn((4,3,S,S))\n\nmodel = unet_swin(img_size=S,size=\"swinv2_small_window16_256\")\nout = model(inp)\n\nprint(out.shape,time.time()-s)\ndel model,out","metadata":{"execution":{"iopub.status.busy":"2022-06-26T06:40:29.640120Z","iopub.execute_input":"2022-06-26T06:40:29.640829Z","iopub.status.idle":"2022-06-26T06:40:34.745541Z","shell.execute_reply.started":"2022-06-26T06:40:29.640787Z","shell.execute_reply":"2022-06-26T06:40:34.744261Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"s = time.time()\nS = 512\ninp = torch.randn((4,3,S,S))\n\nmodel = unet_swin(img_size=S,size=\"swinv2_small_window16_256\")\nout = model(inp)\n\nprint(out.shape,time.time()-s)\ndel model,out","metadata":{"execution":{"iopub.status.busy":"2022-06-26T06:40:34.747712Z","iopub.execute_input":"2022-06-26T06:40:34.748062Z","iopub.status.idle":"2022-06-26T06:40:45.253527Z","shell.execute_reply.started":"2022-06-26T06:40:34.748032Z","shell.execute_reply":"2022-06-26T06:40:45.252045Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"s = time.time()\nS = 512\ninp = torch.randn((4,3,S,S))\nmodel = UperNet_swin(img_size=S,size=\"swinv2_base_window16_256\")\n\nout = model(inp)\n\nprint(out.shape,time.time()-s)\ndel model,out","metadata":{"execution":{"iopub.status.busy":"2022-06-26T06:40:45.256166Z","iopub.execute_input":"2022-06-26T06:40:45.256672Z","iopub.status.idle":"2022-06-26T06:41:00.489771Z","shell.execute_reply.started":"2022-06-26T06:40:45.256634Z","shell.execute_reply":"2022-06-26T06:41:00.488426Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"s = time.time()\nS = 256\ninp = torch.randn((4,3,S,S))\nmodel = UperNet_swin(img_size=S,size=\"swinv2_small_window16_256\")\n\nout = model(inp)\n\nprint(out.shape,time.time()-s)\ndel model,out","metadata":{"execution":{"iopub.status.busy":"2022-06-26T06:41:00.492136Z","iopub.execute_input":"2022-06-26T06:41:00.492487Z","iopub.status.idle":"2022-06-26T06:41:03.316612Z","shell.execute_reply.started":"2022-06-26T06:41:00.492457Z","shell.execute_reply":"2022-06-26T06:41:03.315651Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import segmentation_models_pytorch as smp","metadata":{"execution":{"iopub.status.busy":"2022-06-26T06:43:37.056028Z","iopub.execute_input":"2022-06-26T06:43:37.056459Z","iopub.status.idle":"2022-06-26T06:43:37.062822Z","shell.execute_reply.started":"2022-06-26T06:43:37.056423Z","shell.execute_reply":"2022-06-26T06:43:37.061335Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}